Hyperspectral Anomaly Detection Based on a Beta Wavelet Graph Neural Network

In this article, graph neural networks are applied to hyperspectral anomaly detection, treating each pixel as a node, the correlations between pixels as edges, and the spectral features of each pixel as node attributes, thus fully utilizing both spatial and spectral information in hyperspectral data. Due to the phenomenon of “right-shifted” spectral energy caused by the presence of anomalous targets, the beta wavelet graph neural network detector is proposed in this article; it takes advantage of this phenomenon using beta graph wavelets to generate an efficient, localized bandpass filter to accurately capture anomalous targets. Experimental results show that the proposed method has high accuracy and robustness in hyperspectral anomaly detection.